Environmental and Occupational Health Sciences Institute
Rutgers University Community RepositoryNot a member yet
58345 research outputs found
Sort by
The effects of hormonal contraception on cardiovascular function, somatic experiences, and behavioral patterns of college women: application of the allostatic load theory and somatic marker hypothesis
No full textCardiovascular disease (CVD) is the foremost cause of death among women in the United States. While rates of CVD mortality in women over the age of 65 have been on the decline, incidence and prognosis in women younger than 55 have not improved. This highlights a timely need to understand and detect early alterations in cardiovascular function of young women. This dissertation proposed a framework that integrates concepts from the allostatic load theory and somatic marker hypothesis to identify early risk or protective factors of CVD in young women with a specific focus on the role of hormonal contraception (HC), a known physiological and psychological stressor. Aim 1 leveraged novel breathing tasks of paced sighing and slow-paced breathing to evaluate differences in cardiovascular function between naturally cycling (NC) women and women using HC. Results showed that women using HC had elevated resting systolic and diastolic blood pressures, driven by increased stroke volume and shortened pulse transit time relative to NC women. Despite resting differences, both groups responded similarly to breathing challenges. Alterations in resting measures may indicate early cardiovascular changes associated with HC use. Aim 2 created the Women’s Somatic Embodiment Inventory (WSEI), a novel measure of young women’s subjective somatic states, bodily experiences, and physical perceptions that can be used to assess the role of somatic experiences on behavior, and thus, CVD risk. The final WSEI contains nine unpleasant and four pleasant subscales that reflect two unipolar dimensions. NC women and women using HC scored similarly across most subscales, but women using HC scored higher on the Activated & Attractive subscale. Finally, Aim 3 used latent profile analysis to identify homogeneous subgroups of exercise and alcohol behaviors in college women and evaluated the relationship between profile and markers of mental and somatic health. Four profiles were identified that reflected a variety of exercise and drinking patterns, but did not support a direct exercise-alcohol relationship. Rather, results suggested individual differences (e.g., HC status and premenstrual disorders) and exercise type (e.g., resistance versus aerobic) likely impact the exercise-alcohol relationship. Resistance exercise appears to confer mental health benefits, but more research is needed to understand what facets of resistance exercise are beneficial. Overall, findings from all three aims support the utility of the proposed framework as a way to study women’s health issues. HC use appears to have mild effects across cardiovascular, somatic, and behavioral domains that are influenced by individual differences. More research is needed to understand the effects of HC on women’s health from a holistic perspective, especially relative to CVD etiology, progression, and prognosis.Ph.D.Includes bibliographical reference
Multiscale tests for point processes and longitudinal networks
No full textWe propose a new testing framework applicable to both the two-sample problem on point processes and the community detection problem on rectangular arrays of point processes, which we refer to as longitudinal networks; the latter problem is useful in situations where we observe interactions among a group of individuals over time. Our framework is based on a multiscale discretization scheme that consider not just the global null but also a collection of nulls local to small regions in the domain; in the two-sample problem, the local rejections tell us where the intensity functions differ and in the longitudinal network problem, the local rejections tell us when the community structure is most salient. We provide theoretical analysis for the two-sample problem and show that our method has minimax optimal power under a Holder continuity condition. We provide extensive simulation and real data analysis demonstrating the practicality of our proposed method. In the first chapter, we delve into the two-sample problem and the details of our testing procedure. We provide theoretical analysis for the two-sample test and show that our method has minimax optimal power.
We define the concept of longitudinal networks and elucidate our approach to testing for community structure within them in the Chapter II. We outline our tests across three distinct settings: symmetric networks with homogeneous baseline rates, asymmetric networks with homogeneous baseline rates, and degree-corrected networks with heterogeneous baseline rates.
For both problems, we provide extensive simulation and real data analysis demonstrating the practicality of our proposed method.Ph.D.Includes bibliographical reference
Learning and control in trustworthy and responsible artificial intelligence for cyber-physical systems
No full textWith breakthroughs in technology and excellence in robustness, artificial intelligence (AI) has gradually reshaped the way of information interaction and has been challenging traditional cyber-physical systems (CPS) research. Nourished by enormous data, AI and data-driven machine learning have been emerging with CPS ideas for better capability and adaptability in automation and autonomy. Yet, concerns about the emerging ideas have been raised on dependability and accountability. The data-driven models have been proven to over-reliance on training data, which exposes unreliable generalization and is sensitive to data disturbances. It is true that a larger and broader training dataset could tend to a better model, however, gathering such a dataset is a time-consuming, labor-intensive, and cost-demanding progress with threats to individual privacy. Therefore, it is inevitable to find ethical and robust alternative solutions, among which an interest is in trustworthy and responsible AI. This dissertation explores the learning and control in trustworthy and responsible AI for CPS from a control perspective. In particular, this dissertation intends to improve the performance of data-driven learning on non-cooperative facial-credentialed security authentication within complex backgrounds.
Limited by the physical and cyber components, the quality of visual data within CPS is generally worse than that in datasets. Typically, the collection of datasets for facial-credentialed security authentication tends to address variation in pose, size, illumination, and occlusion but not in facial quality (scales). However, facial quality is an inevitable factor affecting non-cooperative facial-credentialed security authentication. Feature Super-Resolution Face Recognition Net (FSRFR-Net) is designed to balance the quality-effect on facial-credentialed recognition. It utilizes the trained data-driven models, which are convolutional neural networks trained on high-quality faces, and improves their performance on low-quality faces with a minimum change, which is to insert a multi-branch feature super resolution module, in the structure. Experiments on different datasets present consistency in performance improvement on low-quality recognition while preserving high-quality recognition performance. The performance improvement difference among datasets demonstrates the high correlation between the larger face region and better recognition performance.
Following that observation, an anchor-free duo-agent reinforcement learning model is proposed for multi-scale (multi-quality) face detection. It measures the image quality caused influence from the weighted combination of channel-wise feature maps. The design of duo-agent allows the freedom in the pixel-size (resolution) of faces while preserving computational simplicity. The detection goal is to maximize the facial region while minimize the background region. In addition, the duo-agent reinforcement learning design transforms the face detection problem into a classic global vs. local extrema regression problem in the 3D feature space. Consequentially, an intention is to find a stable and convergence solution to the regression problem.
In order to solve the stochastic discrete-time model-free duo-agent regression problem, the convergence efficiency and stability of reinforcement learning approaches in CPS have been explored from the control theory perspective. A mild condition has been proven applicability to policy iteration approach of zero-sum differential games and H∞ robust optimal control. A Simultaneous Online Tracking and Planning (SOTP) Navigation system of autonomous tracked vehicles for leader-follower formation is proposed to examine the optimal strategies obtained through policy iteration approach of multi-agent linear-quadratic optimal control problems. The SOTP framework is analyzed and designed through a skid-steering dynamic model with optimization of low-level controls (agents) of autonomous tracked vehicles for the leader-follower formation. Simulation results have shown the efficacy and robustness of the SOTP framework while the freedom of trajectory curvature and formation posture are reserved without the virtual followers/leaders.Ph.D.Includes bibliographical reference
Dietary animal fat disrupts gut microbiota and aggravates Graft-versus-host disease following hematopoietic stem cell transfer
No full textThe gut microbiome and dietary regimen are two modifiable variables that have been widely investigated as factors that can impact Graft-versus-host disease (GVHD) following allogeneic bone marrow transplant. The rise in fat content has been linked to a reduction in gut microbiome diversity and an aggravation of GVHD. However, the impact of the quality of dietary fat on both the gut microbiota and GVHD outcome remains elusive. Here, we developed two rodent diets with moderate fat, ensuring equivalent macronutrient composition while varying only in fat type—animal-derived versus plant-derived fat. We found that mice fed the animal fat diet (AFD) compared to a plant fed diet (PFD) exhibited worse outcomes in a model of chronic GVHD. The mice fed an AFD had an elevated serum cytokine response, more severe skin inflammation and more intestinal immune dysregulation, compared to mice fed a PFD. While both diets caused reductions of gut microbial diversity, the diets induced discrete gut microbial structures. Further, we identified two guilds consisting of 18 amplicon sequence variants (ASVs) that distinguished between AFD and PFD-fed mice and may contribute to disease severity in AFD group. Our data show that even short-term feeding of diets that are equal in macronutrient content but different in fat source can promote different gut microbiomes, alter intestinal homeostasis, and drive differential Scl-cGVHD outcomes.Ph.D.Includes bibliographical reference
Natural function and perceptual content
No full textThis dissertation develops two interrelated theories: A theory of natural function, which tells us in virtue of what a particular biological item has the function it has; and a theory of perceptual content, which tell us in virtue of what a particular perceptual state represents what it does (so as to allow for misrepresentation). The two theories are interconnected insofar as I use my theory of natural function to ground my theory of perceptal content. The dissertation sets out to resolve a number of long standing issues that have plagued previous theories of perceptual content. In particular the dissertation, when taken as a package deal, endorses a theory of peceptual content that:
(i) Does not face issues concerning swamp-people and related phenomena (chapters 1 and 3).
(ii) Avoids a number of content indeterminacy issues (chapter 3).
(iii) And, allows for cases of reliable/systematic misperception (chapter 4).Ph.D.Includes bibliographical reference
In vivo evolution of Lactobacillus rhamnosus results in prolonged persistence in the digestive tract
No full textLactobacillus rhamnosus, is a renowned probiotic organism found in many food supplements. This thesis research explored in-vivo evolution as a method to generate bacterial strains with longer bioavailability in the intestinal tract. Lactobacillus rhamnosus was autologously gavaged to mice multiple times to form two evolutionary variants of the bacteria after repeated passage through the intestinal tract. A significantly longer retention time (nearly 3x) and a slower elimination rate of the bacteria in the mouse gut was observed with each evolution. The evolutionary strains were further characterized for improved traits of gut retention such as bile salt resistance, epithelial cell binding, and genetic alterations to understand potential mechanistic hypotheses. Finally, a series of heterologous gavages were performed to determine if the increased retention of the evolutionary variants were because of animal specific host adaptations. Similar results were seen following these gavages, with longer retention and slower elimination observed between the wild type and the evolutionary isolates. Based on these findings, in-vivo evolution shows promise as a technique to generate probiotic strains with improved traits for gut retention as compared to the wild type.M.S.Includes bibliographical reference
Efficacy of kernel bypass for high-speed regular expression matching
No full textRegular expression matching is a critical component of various software domains, par-ticularly in network security where it is used to identify malicious traffic efficiently and with a lot of flexibility. Traditional regular expression engines are now struggling to cope up with the ever increasing network traffic volume. There have been many works oriented towards boosting the performance of regular expression matching through mechanisms such as computation offload to the hardware, and kernel bypass through DPDK. This thesis explores the potential use of Hyperscan regular expression matching engine with in-kernel packet access acceleration techniques such as eXpress Data Path (XDP) and shared memory buffers since it does not enlarge the trusted computing base and can achieve the performance benefits of kernel bypass with commodity software.
A system is created where XDP programs are used to directly write the packets into BPF ring buffer or the BPF perf buffer, which enables faster data access. Through XDP the conventional kernel network stack is avoided along with an expensive allocation of socket buffer. A supporting user space application is developed that polls from this shared memory and feeds the content of the packets to the Hyperscan program.
Experimental results show that this system achieves performance sufficient to hit the maximum rate at which hyperscan can process a given payload against a standard set of regular expressions, regardless of the type of buffer used.M.S.Includes bibliographical reference
A touchscreen interface and perception for a socially cognizant campus robot guide
No full textA robotic campus guide was designed, built, and tested. Socially cognizant designprinciples, including the assembly of an interdisciplinary team, engaging with community members, and consideration of unintended consequences, were practiced. The result of these efforts was a robotic system that could interact with a user via a speech and touchscreen interface, and either provides instructions on reaching a location on campus, or guide the user to that location. When guiding the user, the robot detects bystanding people and accounts for them in its motion plan. The system performed reliably during evaluation, although several improvements are envisioned to the hardware and software to make the system more usable and effective. In the context of the Project I have worked on the development of the Touchscreen Interface and Perception Module and its Integration with the Navigation and SLAM nodes.M.S.Includes bibliographical reference
Uncertainty measurement in high-dimensional statistics
No full textWe construct confidence regions under various high-dimensional regression models, including 2-layer neural network estimates, Huber’s robust regression, and shape-constrained regression, highlighting isotonic regression as an example.
In Chapter 2, we study two-layers Neural Networks (NN), where the first layer contains random weights, and the second layer is trained using Ridge regularization. We establish asymptotic distribution results for this 2-layers NN model in the regime where the ratios p n and d n have finite limits, n is the sample size, p is the ambient dimension and d is the width of the first layer. We show that a weighted average of the derivatives of the trained NN at the observed data is asymptotically normal, in a setting with Lipschitz activation functions in a linear regression response with Gaussian features under possibly non-linear perturbations. We then leverage this asymptotic normality result to construct confidence intervals (CIs) for single components of the unknown regression vector. It shows that the double-descent phenomenon occurs in terms of the length of the CIs, with the length increasing and then decreasing as d n ↗ +∞ for certain fixed values of p n.
In Chapter 3, we develop asymptotic normality results for individual coordinates of robust M-estimators with convex penalty in high-dimensions, where the dimension p is at most of the same order as the sample size n, i.e, p/n ≤ γ for some fixed constant γ > 0. The asymptotic normality requires a bias correction and holds for most coordinates of the M-estimator for a large class of loss functions including the Huber loss and its smoothed versions regularized with a strongly convex penalty. The asymptotic variance that characterizes the width of the resulting confidence intervals is estimated with data-driven quantities. This estimate of the variance adapts automatically to low (p/n → 0) or high (p/n ≤ γ) dimensions and does not involve the proximal operators seen in previous works on asymptotic normality of M-estimators. For the Huber loss, the estimated variance has a simple expression involving an effective degrees-of-freedom as well as an effective sample size. The case of the Huber loss with Elastic-Net penalty is studied in details and a simulation study confirms the theoretical findings.
In Chapter 4, we construct optimal confidence balls for the shape restricted Gaussian sequence models, where the mean vector μ to estimate is assumed to be constrained in a closed convex subset K in Rn. Interested in the confidence balls centered at the MLE restricted in K and the optimal data-driven radius among all possible choices, we show that if the data-driven radius is taken to be the square root of the divergence of the MLE up to a constant, the confidence balls are honest but optimal in several regimes. One criterion of the optimality lies in the case when the noise level goes to 0, as is called the “low sigma limit” case. Provided that K is a closed convex subcone of Rn with statistical dimension δ(K) → +∞, the confidence balls are adaptive over conical sub-models of K and are optimal if the signal-to-noise level is extremely high or low. We study the univariate isotonic regression model as an example.Ph.D.Includes bibliographical reference
Axonal collateralization of basolateral amygdala neurons
No full textIt was proposed that basolateral amygdala (BLA) neurons projecting to differentareas encode distinct types of information, allowing them to generate contrasting behaviors. Opposite to this view, other findings indicate that BLA neurons simultaneously encode a heterogenous mixture of sensory and behavioral variables, which might not segregate as a function of their projection sites given that a substantial proportion of BLA neurons send branching axons to multiple sites. To determine which of these views is correct, I recorded BLA neurons projecting to nucleus accumbens (nAc) or the medial prefrontal cortex (mPFC) during a task that requires rats to emit different appetitive or aversive behaviors based on the location of light stimuli on each trial. I found that BLA neurons projecting to distinct sites encode similar types of information. Next, I characterized the axon collateralization patterns of BLA neurons, about which very little was known. Using a viral tracing strategy that allows visualization of the axon collaterals contributed by different projection-defined BLA subpopulations, I found that BLA neurons projecting to the mPFC, anterior insula, nAc, or dorsal striatum (but not the ventral hippocampus or ventromedial hypothalamus) send massive collateral projections to a common array of subcortical structures. Last, I corroborated the existence of this profuse axonal collateralization using multiple fluorescent retrograde tracers.Ph.D.Includes bibliographical reference