22689 research outputs found
Sort by
Non-transportable Statistical Calibration of a Network of Low-Cost NO2 Sensors
No full textLow-cost sensors have gained popularity as a feasible option for monitoring air pollution
due to their ability to provide high resolutions in both space and time. However,
the data quality from these sensors can be unstable and requires calibration and
validation, typically using measurements from regulatory-grade sensors as a reference.
The availability of such data is often limited owing to the high deployment cost. The
sensor-specific biases between each low-cost unit further restrict the direct transport
of an established calibration model for a specific sensor to other units. This study
presents the evaluation of a proxy reference approach to the statistical calibration of
a nitrogen dioxide (NO2) low-cost sensor operating in Baltimore, MD. Three years of
collocated data at Oldtown were used to establish a calibration model with regulatory-grade
sensor data of Essex as a proxy reference, testing different subset approaches,
calibration equations, and model framework. The chosen best non-transportable
calibration strategy shows superior performance compared to alternation calibration
approaches, achieving an hourly root-mean-squared-error (RMSE) of 4.93 ppb. The
close performance between the proxy reference-trained model and the true reference-trained
model further indicates small information loss when the chosen calibration
strategy is employed. The high agreement to the regulatory-grade measurements and
the tolerable loss in calibration performance provides a promising approach for the
statistical calibration of low-cost sensors in a non-collocated setting
Robust Speaker Recognition using Perceptual and Adversarial Speech Enhancement
No full textIn Automatic Speaker Verification (ASV), we determine whether the speaker in the test utterance is identical to the previously enrolled speaker. Deep learning has significantly improved ASV performance. However, it is still susceptible to external disturbances and domain mismatches. A standard solution is data augmentation, i.e., adding noise and reverberations in training data. We focus on developing pre-processing solutions that can be integrated with existing pipelines and advance empirical performance on state-of-the-art systems. For this, we pursue deep learning-based speech enhancement and develop solutions equipped with denoising, domain adaptation, and bandwidth extension (BWE).
Existing speech enhancement solutions often lead to degradation in ASV performance, partly due to loss of speaker information.
We propose using perceptual/deep features that leverage pre-trained models to handle this. We also prove the effectiveness of our denoiser by showing that it complements the missing noise class in the x-vector (training) data augmentation through ablation studies.
We also improve the training data for telephony speaker verification, where it is a common practice to downsample higher-bandwidth microphone speech to lower sampling frequency and apply telephone codecs. We propose to replace this by learning a mapping using a deep feature-based CycleGAN. Our novel technique improves training data and complements the prior techniques, including data augmentation.
To handle bandwidth mismatch, we pursue BWE with time-domain architectures. We develop competent Generative Adversarial Networks (GAN): supervised (conditional GAN) and unsupervised (CycleGAN). Our findings indicate that unsupervised learning can give close performance to supervised performance. We also pursue joint learning BWE schemes with domain adaptation.
Finally, with our proposed Self-FiLM scheme, we leverage self-supervised representations to guide BWE models better in unknown environments.
In conclusion, we provide evidence that speech enhancement can be used as a pre-processor for improving ASV. By testing on real data acquired from Speaker Recognition Evaluation challenges, we demonstrate the effectiveness of speaker-identity preserving denoisers. We also study the effectiveness of domain adaptation and Self-Supervised Learning to improve bandwidth extension. Our work opens research into a joint investigation of enhancement-related problems and better generative models to assist the x-vector
IDENTIFYING A NOVEL “DARK-YOLK” GENE FOR ITS ROLE IN LIPOPROTEIN BIOGENESIS IN DANIO RERIO
No full textLipids are essential nutrients for the human body- they are needed for energy, act as chemical messenger, are signaling molecules and are integral component of cell membranes. Additionally, they aid with vitamin absorption, keep us warm, and protect vital organs from damage. Triglycerides (fat) are lipids composed of one molecule of glycerol bound to three fatty acids. In order to transport hydrophobic triglycerides through the aqueous circulatory system from where they are absorbed/synthesized to where they are needed for energy/storage, triglycerides are packaged into Apolipoprotein B-containing lipoproteins (B-lps). We still have an incomplete understanding of the genetic factors that influence B-lp biogenesis and metabolism. Increasing our understanding in this field is essential as overproduction of B-lp is associated with atherosclerosis. B-lp processing is conserved, thus is an essential process in vertebrates and can be studied in various model organisms. Zebrafish is a vertebrate model that can be used for a forward genetic screen and can undergo efficient genome editing. A large-scale forward genetic screen in zebrafish for a phenotype associated with abnormal B-lp biosynthesis identified 27 mutant families. The aim of my project is to characterize one of the novel mutants, Mutant 5, which, in preliminary studies, was shown to have decreased B-lp levels. I specifically worked on identifying the causal mutation for the Mutant 5 phenotype. I used PCR to identify genomic regions of interest. Using CRISPR/cas9 genome editing, I targeted candidate genes to test to see if they phenocopied Mutant 5. This data led to the identification of slc3a2a, a gene previously unlinked to B-lp metabolism that is conserved in humans. This work provides new insight into the molecular players involved in regulation of lipid flux across species
From Farmers to Scientists: The West Riding Pauper Lunatic Asylum as a Research Institution
No full textThe West Riding Pauper Lunatic Asylum has a reputation as a center of neurological and psychiatric research in Victorian England. That reputation rests on its being the site of the cerebral localization work of Sir David Ferrier (1873). While Ferrier’s work was undoubtably influential, the work done at the Asylum was wide-ranging in topic, as the asylum physicians struggled to elucidate the etiology of mental disease. This thesis argues that the asylum did not become a research center for mental disease and brain science by happenstance. I show that this evolution was the product of the careful plan of Sir James Crichton-Browne, the medical superintendent from 1866 through late 1865. Careful analysis of primary sources including the institution’s Medical Reports, the asylum’s reports to its supervisory committee, a prosopography of the researchers at the asylum, and the research trajectories of the careers of those researchers show that Crichton-Browne very carefully planned and developed a research institution within his asylum. Establishment of the West Riding Asylum as a research institution, carefully developed by Crichton-Browne, played a role in the development of neurology and psychiatry as a medical specialty in Britain
UNSUPERVISED SEGMENTAL MODELING OF SPEECH FOR LOW RESOURCE APPLICATIONS
No full textVoice-enabled interfaces for human-machine interaction have made significant progress in recent years. Most of the success can be attributed to deep neural networks trained on thousands of hours of transcribed data. However, vast amounts of labeled data are not available for most spoken languages worldwide, e.g., regional languages. Here we explore alternate techniques that can learn directly from data without any or minimal manual transcriptions.
The representation techniques employed to characterize the speech signal dictate the performance of unsupervised systems. Self-supervised methods such as Contrastive Predictive Coding (CPC) have emerged as a promising technique for representation learning from unlabeled speech data. Based on the observation that the acoustic information, e.g., phones, changes slower than the feature extraction rate in CPC, we propose regularization techniques that impose slowness constraints on the features. First, we propose two regularization techniques: Self-expressing constraint and Left-or-Right regularization. Our modifications outperform the baseline CPC in monolingual, cross-lingual, or multilingual settings on the ABX and linear phone classification benchmarks.
However, CPC or our modifications mainly look at the audio signal's structure at the frame level. The speech structure exists beyond the frame level, i.e., at the phone level or even higher. We propose a segmental contrastive predictive coding (SCPC) framework to learn from the signal structure at both the frame and phone levels. SCPC is a hierarchical model with three stages trained in an end-to-end manner. In the first stage, the model predicts future feature frames and extracts frame-level representation from the raw waveform. In the second stage, a differentiable boundary detector finds variable-length segments. In the last stage, the model predicts future segments to learn segment representations. Experiments show that our model outperforms existing phone and word segmentation methods on TIMIT and Buckeye datasets.
In the last part, we explore knowledge distillation from text encoders (e.g., Roberta) to speech encoders in an unsupervised manner in a multimodal setting. Text encoders operate at the sub-word level, while speech encoders operate at a much smaller scale, i.e., frames. Our segmental framework allows us to downsample frames and generate sub-words. SCPC enables us to leverage pretrained text encoders in an audio-visual and audio-only setting. We show significant performance improvements on the audio-image retrieval and semantic similarity task
A Qualitative Exploration of Funding Drivers for Women Entrepreneurs in Health Care
No full textIntroduction: Women are the predominant force in the United States health care system but only receive a small fraction of the venture capital funding. This imbalance calls for an industry-specific investigation of the funding drivers and entrepreneurship process.
Methods: 29 semi-structured interviews were conducted with key informants (x13) and entrepreneurs (x16) with experiences directly related to new venture development in the U.S. health care industry. The Framework Method was utilized to complete a thematic analysis of the data, which was informed by historical themes from entrepreneurship literature, feminist theory, and intersectionality.
Results: Several drivers of lower entrepreneurial funding for women in health care were identified and help explain the failure of venture capital to deploy resources rationally within this specific ecosystem. These include a male dominated investor base that struggles to appreciate a portion of the discovery and creation opportunities, problematic social norms and caregiver disparities that still permeate working environments, financing inequities that include narrow standards for investment, and messaging that positions women as unprepared and unincluded.
Conclusion: This study recognizes their achievements while also presenting the intersection of women entrepreneurship and health care as a uniquely important area of research from which to analyze funding drivers, the entrepreneurship process, and specific policy implications
Automating the Analysis and Improvement of Dynamic Programming Algorithms with Applications to Natural Language Processing
No full textThis thesis develops a system for automatically analyzing and improving dynamic programs, such as those that have driven progress in natural language processing and computer science, more generally, for decades. Finding a correct program with the optimal asymptotic runtime can be unintuitive, time-consuming, and error-prone. This thesis aims to automate this laborious process. To this end, we develop an approach based on
(1) a high-level, domain-specific language called Dyna for concisely specifying dynamic programs
(2) a general-purpose solver to efficiently execute these programs
(3) a static analysis system that provides type analysis and worst-case time/space complexity analyses
(4) a rich collection of meaning-preserving transformations to programs, which systematizes the repeated insights of numerous authors when speeding up algorithms in the literature
(5) a search algorithm for identifying a good sequence of transformations that reduce the runtime complexity given an initial, correct program
We show that, in practice, automated search—like the mental search performed by human programmers—can find substantial improvements to the initial program. Empirically, we show that many speed-ups described in the NLP literature could have been discovered automatically by our system. We provide a freely available prototype system at https://github.com/timvieira/dyna-p
Improving College Readiness Through Arts, Athletics, and Family Engagement
No full textAmerican students are not graduating high school as college-ready learners. To achieve college readiness, students can learn to become well-rounded by studying academics, arts, and athletics in high school. Unfortunately, recent federal and state policies discourage this option and do not promote extracurricular participation in American high schools. A mixed-methods study examined college readiness and extracurricular participation from 10 first-year college students' perspectives to find an avenue for intervention. The findings indicated that students' family members significantly impact their extracurricular participation and, therefore, their college readiness. Additional research shows that schools can partner with families to effectively use their influence to jointly impact students' educational choices and trajectories. This paper culminates with a partnership guide for school administrators to use to partner with families. A School Administrator’s Guide to Improving College Readiness Through Arts, Athletics, and Family Engagement provides schools with the blueprint for reevaluating their approach to students’ extracurricular participation. The guide and its virtual companion at CollegeReadyGrads.com feature a step-by-step procedure for implementation that helps schools foster a partnership with families based on the common goal of increasing college readiness
Characterizing the Content and Context of Celebrity Mental Health Discourse on Instagram: An Arts-Based Dissertation
No full textBackground: Mental illness is a leading contributor to morbidity and mortality, and disproportionately affects young people. Young celebrities have increasingly discussed their mental health on popular social media platforms such as Instagram. However, little research has been done to understand these disclosures. Celebrities may act as role models for youth; the messages that celebrities post to their millions of followers have significant impact. The study aimed to answer these research questions: 1) What are the key themes present in mental health-related posts on Instagram by top young adult celebrities?, 2) What functional types of purpose do the mental health-related posts serve?, 3) What contextual factors/events prompted mental health-related posts?
Methods: This qualitative multi-case study examined Instagram accounts for 19 of the most popular young celebrities who had previously self-disclosed having a mental illness. Instagram posts from January 1, 2019-December 31, 2020 were examined and mental health posts were thematically analyzed. The study used a novel application of Goffman’s dramaturgical theory and an arts-based format through researcher-created simulated Instagram posts.
Results: Analysis involved reviewing 11,180 Instagram posts, of which 161 mental health posts were identified. Thematic analysis revealed two main themes: identity management and frontstage/backstage duality. Mental health posts were categorized into five types of functions they served: disclosure, advocacy, educational, motivational, and self-promotion. Analysis of the context found four main social contextual events coinciding with mental health posts and linked to their content: the COVID-19 pandemic, murder of George Floyd, 2020 US election, and death of Kobe Bryant.
Conclusions: The study examined social media through a qualitative, holistic lens and provides a foundation for future research that examines the impact that celebrities have on mental health discourse and the potential role that Instagram plays in being a platform for subversive mental health discourse and intervention. The findings are presented in an arts-based format using Instagram posts to represent the study in a way that will be engaging for persons the study is intended to benefit. This innovative method demonstrates the potential for social media to be used for arts-based research in the field of public health
TOWARD UNDERSTANDING THE ALTERNATIVE ACTIVITIES AND DONOR-DEPENDENT MECHANISTIC CONTROL OF BACTERIAL DXP SYNTHASE
No full textThe bacterial enzyme 1-deoxy-D-xylulose 5-phosphate synthase (DXPS) catalyzes the thiamine
diphosphate (ThDP)-dependent formation of DXP from pyruvate and D-glyceraldehyde 3-phosphate (D-GAP). DXP is a branch-point metabolite, as it is required for isoprenoid biosynthesis and ThDP and pyridoxal phosphate cofactor biosynthesis. Due to the essential role of DXPS in bacterial metabolism and its demonstrated catalytic promiscuity and conformational diversity, we hypothesize DXPS could contribute to bacterial adaptations in an ever-changing metabolic landscape during infection. If this is true, the potential multifunctional roles of DXPS could be targetable for antibacterial drug design. Studies by the Freel Meyers lab have elucidated the relaxed specificity toward DXPS acceptor substrates, but previous reports have suggested a strict requirement for pyruvate as the donor substrate in the presence of the natural acceptor substrate, D-GAP. Here, we show that DXPS can indeed utilize alternative ketoacid, aldehyde or ketose phosphate donor substrates, and propose that donor substrate usage may be dictated by the nature and availability of acceptor substrate. The existence of alternative DXPS donor/acceptor substrate pairs sets a precedent for unique donor/acceptor combinations which could have important implications for DXP synthase function. In fact, DXPS can catalyze the formation of a dihydroxyethyl (DHE)ThDP carbanion intermediate, a physiological intermediate for other ThDP-dependent enzymes. Formation of DHEThDP from a ketoacid or aldehyde donor substrate, via decarboxylation or deprotonation, respectively, is described in Chapter 2. We also demonstrated that DXPS can utilize D-xylulose 5-phosphate (X5P), the physiological substrate for transketolase, thereby producing DHEThDP through a third mechanism, D-GAP elimination (Chapter 3). In this study, we explored the fate of DHEThDP formed from X5P on DXPS, including its stability and acceptor substrate specificity and found DXPS exhibits TK-like activity, producing D-fructose 6-phosphate from X5P and D-erythrose 4-phosphate. We hypothesize the alternative activities of DXPS may be controlled by conformational dynamics and discuss the design of a conformational DXPS probe that could be used to interrogate conformational dynamics in this system (Chapter 4). In total, the findings described here provide evidence for the rare donor-substrate flexibility of DXPS, a trait that could be exploited for biocatalysis development and potentially for drug discovery