48670 research outputs found
Sort by
Dialect-specific Acoustic Correlates of Stress in Spanish: The Role of Vowel Compression and Syllable Structure
Get PDFThe present paper is part of a wider project examining vowel compression and its impact on the phonetic signalling of stress across dialects of Spanish. Here within, we compare vowel compression effects in spontaneous Southern Chilean Spanish to previous findings from continuously read Altiplateau Mexican Spanish. Results show that, in Southern Chilean Spanish, vowels are shortened in CVC and CCV syllables irrespective of stress; although unstressed vowels are shorter than stressed, onset and coda-driven compression effects are visible on all vowels. Qualitative results show that stress-driven differences in vowel height are visible on /o/ and /a/ in open but not closed syllables: stressed vowels are lower in the vowel space. Conversely, results from Altiplateau Mexican Spanish showed that whilst unstressed vowels in CVC syllables were shortened and centralised, stressed vowels were not. Results therefore support theories that dialect-specific compression effects exist due to dialect-specific phonetic-phonological interactions (Authors under review): in this case, their interaction with stress. We further consider the implications of these variety-specific patterns in the context of debates concerning the dialect-specific nature of stress, arguing that compression effects may have implications on the wider vowel systems and the phonetic way in which stress is signalled across dialects
Directionality Effects and Exceptions in Learning Phonological Alternations
Get PDFThe present study explores learning vowel harmony with exceptions using the artificial language learning paradigm. Participants were exposed to a back/round vowel harmony pattern in which one affix (either prefix or suffix, depending on the condition) alternated between /me/ and /mo/ depending on the phonetic feature of the stem vowels. In Experiment 1, participants were able to learn the behaviors of both alternating and non-alternating affixes, but were more likely to generalize to novel affixes for non-alternating items than alternating items. In Experiment 2, participants were exposed to training data that contained non-alternating affixes in prefix position while alternating affixes were all suffixes, or vice versa. Participants were able to extend the non-alternating affixes to the novel direction, suggesting that participants inferred a non-directional harmony pattern. Overall, the patterns of alternating affixes are harder to learn than patterns of exceptions that do not alternate, which aligns with previous findings supporting a non-alternation bias. Our study raises the question of how biases towards exceptionality and directionality interact in phonological learning
Machine Learning and Quantitative Neuroimaging in Epilepsy and Low Field MRI
No full textMedical imaging plays a key role in the diagnosis and management of neurological disorders. Magnetic resonance imaging (MRI) has proven particularly useful, as it produces high resolution images with excellent tissue contrast, permitting clinicians to identify lesions and select appropriate treatments. However, demand for MRI services has outpaced the availability of qualified experts to operate, maintain, and interpret images from these devices. Radiologists often rely on time-consuming manual analyses, which further limits throughput. Moreover, a large portion of the world’s population cannot currently access MRI, and demand for medical imaging services will continue to increase as healthcare quality improves globally. To address these challenges, we must find innovative ways to automate medical processing and produce lower-cost medical imaging devices. Recent advances in deep learning and low-field MRI hardware offer potential solutions, providing lower-cost methods for processing and collecting images, respectively. This thesis aims to develop and validate lower-cost methods for collecting and interpreting neuroimaging using machine learning algorithms and portable, low-field MRI technology. In the first section, I develop a deep learning algorithm that automatically segments resection cavities in epilepsy surgery patients and quantifies removed tissues. I also compare the impacts of epilepsy surgery on remote brain regions, demonstrating that more selective procedures minimize postoperative cortical thinning. In the second section, I explore and validate clinical applications for a new portable, low-field MRI device. Using open-source imaging and machine learning, I propose a low-cost method for simulating diagnostic performance for novel imaging devices when only sparse data is available. Additionally, I validate device performance in multiple sclerosis by directly comparing the low-field device to standard-of-care imaging using a range of manual and automated analyses. My hope is that machine learning and low-field MRI will increase medical imaging access and improve patient care worldwide
Machine Learning as Tool and Theory for Computational Neuroscience
No full textComputational neuroscience is in the midst of constructing a new framework for understanding the brain based on the ideas and methods of machine learning. This is effort has been encouraged, in part, by recent advances in neural network models. It is also driven by a recognition of the complexity of neural computation and the challenges that this poses for neuroscience’s methods. In this dissertation, I first work to describe these problems of complexity that have prompted a shift in focus. In particular, I develop machine learning tools for neurophysiology that help test whether tuning curves and other statistical models in fact capture the meaning of neural activity. Then, taking up a machine learning framework for understanding, I consider theories about how neural computation emerges from experience. Specifically, I develop hypotheses about the potential learning objectives of sensory plasticity, the potential learning algorithms in the brain, and finally the consequences for sensory representations of learning with such algorithms. These hypotheses pull from advances in several areas of machine learning, including optimization, representation learning, and deep learning theory. Each of these subfields has insights for neuroscience, offering up links for a chain of knowledge about how we learn and think. Together, this dissertation helps to further an understanding of the brain in the lens of machine learning
Advancements in Modeling and Simulating Nonadiabatic Dynamics in Electrochemical Systems
No full textCurrent methods for modeling electron transfer (ET) dynamics in electrochemical systems rely on boundary condition based methods that often obfuscate any nonadiabatic effects. However, these nonadiabatic effects can play critical roles in many key electrochemical systems, such as ET events coupled to proton transfer, or systems where ET occurs between metal adsorbed redox species and the metal. In this dissertation we present new methods and algorithms for modeling and simulating electrochemical systems exhibiting nonadiabatic effects. First, we examine whether electron friction dynamics can be used in lieu of surface hopping generally for nonadiabatic dynamics. We find that electronic friction is only applicable in specific situations, and that in general a broadened classical master equation solved with surface hopping must be applied in the weak molecule-metal coupling limit. Next, we outline a surface hopping approach for modeling sweep voltammetry experiments, and show a technique for mapping multidimensional, diffusional electrochemical systems to reduced, lower dimensional systems. This surface hopping approach is used to study proton coupled electron transfer (PCET), whereby nuclear quantum effects are incorporated through an explicit proton coordinate. Finally, we present a spatially grid-free algorithm for simulating the above electrochemical systems (including systems where inner-sphere effects are included, such as adsorption), and also demonstrate its computational efficacy in comparison to current simulation techniques common in the field of electrochemistry
Statistical Approaches to Reducing Bias and Improving Variance Estimation in the Presence of Covariate and Outcome Measurement Error
No full textLarge epidemiologic studies with self-reported or routinely collected electronic health records (EHR) data are frequently being used as cost-effective ways to conduct clinical research, but these types of data are often prone to measurement error. While large epidemiologic studies play a crucial role in understanding the relationship between risk factors and health outcomes, such as disease incidence, these relationships cannot be properly understood unless methods are developed that reduce the bias caused by errors in both exposure variables and time-to-event outcome variables. Furthermore, variance estimates for outcome model regression parameters can be quite large in the presence of complex error-prone exposures and outcomes, yet strategies to improve variance estimation have been given little attention in the measurement error literature. Throughout this dissertation, we address these gaps in the literature by developing methodology that focuses on (1) reducing the bias that occurs from both error-prone exposures and outcomes in large epidemiologic cohort studies with periodic follow-up, (2) improving statistical efficiency by leveraging error-prone, auxiliary data alongside validated outcome data, and (3) considering alternative, better-behaved variance estimation strategies that may be used when techniques for adjusting for measurement error are applied. In Chapter 2, we present a method that combines an approach for addressing errors in event classification variables with regression calibration, a popular technique for addressing exposure error. This method reduces the bias induced by measurement errors in a discrete time-to-event setting. We apply our method to data from the Women’s Health Initiative (WHI) study to evaluate the association between dietary energy and protein and incident diabetes. Chapter 3 develops an approach for incorporating error-prone, auxiliary data into the analysis of an interval-censored time-to-event outcome. Here, the key goal is to improve statistical efficiency in the estimation of exposure-disease associations. We extend our methodology to handle data from a complex survey design and to be used in conjunction with regression calibration. Using this approach, we assess the association between energy and protein and the risk of diabetes in our motivating study, the Hispanic Community Health Study/Study of Latinos (HCHS/SOL). In Chapter 4, we propose a sandwich variance estimator as an approach for accounting for the uncertainty added by using an estimated exposure when regression calibration is applied to adjust for covariate error. This variance approach broadly applies to other two-stage regression settings. We outline a procedure for easily computing the sandwich in standard software and assess its properties through a numerical study and through illustrative data examples from the WHI and HCHS/SOL studies. Our results show that this method may have advantages over commonly applied, resampling-based variance estimation approaches
How Do Your Genes Smell? Applying Genetic Tools to Understand Olfactory Perception
No full textMany of the major milestones in understanding the olfactory system have arisen from research conducted in non-human animal models through the use of various genetic tools. While mammalian olfaction is largely conserved, there are several key differences in humans that suggest the necessity of studies directed at human olfaction. The human olfactory system has both genotypic variation and phenotypic variation, suggesting that, like in animal research, genetic tools are viable and vital for investigating human olfaction. We first address the basic questions of the human olfactory system, examining whether our assumptions based on animal olfaction are confirmed and attempting to identify mechanisms of the olfactory system that are not well-characterized in humans or other animals. We conducted whole exome sequencing on 52 individuals in ten families with congenital anosmia, an inherited disorder where individuals lack a sense of smell from birth. Through selection of rare, segregating, deleterious variants, we identified 215 genes that are associated with congenital anosmia. These genes are likely to play a role in olfaction, and can therefore serve as a resource for further investigation of the underlying mechanisms of basic human olfaction. We next investigated the relationship between olfactory receptors and odor perception in order to improve our understanding of how odor information is processed at the periphery. Here, we identified two novel associations: a variant in OR51B2 that increases intensity perception of a key component of body odor, and two-linked variants in OR4D6 that predict specific anosmia to the musk compound Galaxolide. Uncovering the relationship between odors and their specific receptors is the first step to understanding how the ~400 olfactory receptors work in combination to code for odor perception. These studies make advances towards a comprehensive understanding of how the human peripheral olfactory system functions and codes olfactory information into odor perception
Beyond Academics: A Correlational Study of Social Emotional Learning Competency Growth and Student Demographic Data
Get PDFA growing body of evidence highlights the relationship between youth attainment of social emotional learning (SEL) competencies and school outcomes such as academic performance, school attendance, school attainment, behavioral problems in school, and persistence of antisocial behavior. A lack of clear diversity in student populations in prior research leads to questions regarding growth in social emotional learning competencies in diverse populations. Given disparities in academic performance between low-income minority students and their affluent peers and the potential impact of SEL on student outcomes, this topic deserves further exploration. The present study explored social emotional learning competency growth in youth investigating the following questions:
To what extent do student demographic and socio-cultural characteristics (gender, age, enrollment in special education, English language learner status, free/reduced lunch status) relate to social emotional learning competency growth?
In what way does a student’s grade level relate to their rate of social emotional learning competency growth?
This correlational study explored SEL competency data collected by a large charter school network via the Panorama Social Emotional Survey, an open-source assessment that measured student social emotional skills and mindsets via student self-report. The survey was conducted twice, once in the fall and again in the spring. Multiple regression analysis and One-way Analysis of Variance (ANOVA) tests were used to answer the research questions. Regression models were tested with three measures of social emotional learning competencies in spring as dependent variables, demographic and socio-cultural characteristics as independent variables, and fall scores included as control variables. Data suggests positive associations between development in all 3 SEL competencies and age. Student grade level, gender, enrollment in special education, enrollment in free and reduced lunch programming, and English language learner status each had significant positive or negative associations with one to two measures of SEL competency growth. Implications for study findings include expanded research into other SEL competencies and associations with socio-cultural and demographic characteristics, development of SEL interventions, and applications of social emotional learning competency training in diverse and historically underrepresented populations
Penn Library\u27s Ms. Codex 1640 - Manipulus florum peritorum. (Video Orientation)
No full texthttps://repository.upenn.edu/sims_video/1189/thumbnail.jp