19304 research outputs found
Sort by
Architectural Design and Optimization of Quantum-Resistant Cryptographic Processors
No full textThe rapidly growing field of quantum computing offers a wide range of improvements over classical computers across various domains. These advantages stem from various properties of Quantum Computers, such as superposition, entanglement, and interference, which enable an exponential speed-up for specific problems when compared to Von Neumann architectures. However, this computational power also poses a critical threat to modern cryptographic systems. Quantum algorithms such as Shor and Grover, both unique to quantum computers, can effectively break widely used cryptographic schemes. A new field, Post Quantum Cryptography (PQC), is quickly evolving facing the threat of quantum computers. PQC seeks to replace current cryptographic methods, such as the Advanced Encryption Standard (AES) and the Rivest-Shamir-Adleman (RSA) cryptosystems, with new algorithms that can withstand Shor’s algorithm and Grover’s Quantum Search Algorithm. This dissertation explores novel hardware implementations of PQC, focusing on optimizing performance, area, and power efficiency. Advances in cryptographic and mathematical primitives are presented, including low-power entropy generation techniques and hardware-optimized modular reduction beyond base-two arithmetic. Contributions extend to power-constrained hardware, demonstrating both a hyper-chaotic encryption and the most power-efficient SPHINCS+ implementation to date. Additionally, an even lower area alternative to SPHINCS+, Ascon-Sign, is introduced, specifically tailored for devices in the Internet of Things (IoT). Moreover, an exploration is conducted on how new techniques can be used to accelerate computationally expensive algorithms, such as Fully Homomorphic Encryption (FHE). A superscalar accelerator for the Residue Number System variant of FHE is proposed, based on a new method of modular reduction, demonstrating an amortized computational throughput of 682 Gbit/s on commercially available FPGAs, significantly improving the efficiency of encrypted computations
Engineered Approaches Coupled with Imaging of the Heart and Vascular System to Improve Predictive Outcomes, Diagnoses, and Delivery
No full textThe cardiovascular system’s complexity and remodeling response to disease pathology highlight the dynamic behavior necessary for the body to adapt or compensate to external stressors. This response by the cardiac and vascular system to its changing environment further complicates our ability to rapidly diagnose disease abnormalities and deliver timely therapeutics to prevent disease progression. Addressing these challenges requires innovative solutions to improve predictive modeling, enhance diagnostic precision, and optimize therapeutic interventions. Computational modeling provides test conditions for simulating changing vascular flow dynamics, tissue mechanics, and mass transport, yielding predictive insights to guide treatment. Concurrently, imaging technologies, such as computed tomography (CT) and magnetic resonance imaging (MRI), which serve as the gold standards for structural and functional clinical assessments, are not always feasible for intraprocedural evaluation. The lack of real-time, intraoperative imaging modalities limits precision and a direct means to confirm treatment efficacy. Beyond diagnostics, improving drug delivery to the heart and vascular system is crucial, as common administration routes suffer from limited retention, reducing long-term therapeutic efficacy. Optimizing drug localization and sustaining therapeutic release within the myocardium is fundamental to advance precision-targeted therapies. This dissertation integrates computational modeling, imaging technologies, and engineered drug delivery strategies to improve predictive outcomes, diagnoses, and therapeutic delivery. A computational modeling approach based on experimentally acquired parameters is described as a preclinical means to characterize coronary anatomy and hemodynamics. Using morphometric analysis and computational fluid dynamics (CFD) derived from micro-computed tomography (micro-CT) vascular reconstructions, a novel portfolio of geometric and hemodynamic descriptors is presented to provide insights into species-specific arterial characteristics. In parallel, to study the remodeling of peripheral and pathological vasculature, a finite element-based transport model, the Kojic Transport Model (KTM), is utilized to predict intratumoral pressure and velocity distributions in solid murine Lewis lung carcinoma (LLC) tumors. Experimentally derived heat maps capturing vascular density, capillary diameter, and permeability are integrated into the computational network, offering a quantitative framework to assess the influence of tumor vascular dynamics on mass transport. With respect to innovative imaging solutions, the present work also investigates the feasibility of photoacoustic-ultrasound (PA-US) imaging for real-time procedural guidance during cardiac interventions. Feasibility of direct epicardial PA-US imaging is demonstrated in healthy and ablated porcine (i.e., intraprocedural lesions were made using radiofrequency ablation (RFA)). Data supports that when compared to conventional US transducer handheld imaging, stand-off imaging, and integration of a custom 3D-printed suction coupler without and with vacuum the mean axial displacements were significantly reduced from 0.89, 0.49, 0.49, versus 0.15 mm, respectively (p-value ≤8.65e-7). Comparing the coupler without suction to that with suction, applied lateral interframe displacement reduced from 4.58 to 2.01 mm, respectively (p-value = 5.07e-23) with overall reduced cardiac tissue motion. PA-US based measurements of RFA lesion dimensions matched (<0.7 mm error) gross pathologic assessment, yielding in vivo feasibility data to provide intraoperative lesion dimensions. Moving towards treatment approaches, an intramyocardial in situ delivery strategy is evaluated in live porcine models to locally distribute gold nanoshells through incorporation in a calcium alginate hydrogel (GNS gel). Both transepicardial and transendocardial injections demonstrated successful myocardial retention of the GNS gel for up to 3 hours in acute studies, with 3D reconstruction and volumetric quantification performed based on CT images. By integrating in silico, in vitro, and in vivo methodologies, this dissertation advances computational modeling, diagnostic imaging, and precision-targeted therapies, accelerating use of these preclinical approaches towards clinical cardiac and vascular care
Towards Robust Person Re-Identification: Learning Invariant Features Under Clothing Changes and Occlusions
No full textPerson Re-Identification (Re-ID) involves matching the same person in a non-overlapping camera system. This is a crucial task in computer vision with applications spanning search-and-rescue, safety, and surveillance. However, achieving robust performance in real-world, unconstrained environments remains a significant challenge due to variations in clothing and occlusions. This research addresses these “in-the-wild” conditions by proposing novel methods for learning robust and invariant feature representations capable of generalizing across these challenging scenarios. To tackle Cloth-Changing Re-ID (CCRe-ID), we first introduce a multi-modality framework that incorporates contrastive viewpoint-aware shape learning, enhancing shape robustness against viewpoint variations. Next, we propose a contrastive clothing and pose generation approach using adversarial training to augment identity representations with diverse clothing styles while maintaining pose consistency. Extending to video-based CCRe-ID (VCCRe-ID), we introduce a temporal 3D shape modeling framework that extracts shape dynamics for improved recognition and a spatial-temporal graph attention network that explicitly models gait as a complementary feature to shape and appearance. Finally, we propose a cross-modality complementary learning strategy, which leverages silhouette masks and cross-attention mechanisms to adaptively fuse information from appearance and shape, ensuring robustness under viewpoint variations. For Occluded Re-ID (OCCRe-ID), we first present an attention-based cross-modality learning framework that leverages occlusion-aware input representations to enhance feature robustness. We then extend this approach to a Cross-Attention Vision Transformer (ViT), enabling fine-grained local feature extraction that improves Re-ID performance under occlusions and clothing changes. To explicitly model occlusion types, we propose an occlusion-aware appearance and shape reasoning framework, incorporating occlusion-type information to guide the Re-ID model in distinguishing between occluded and non-occluded features. Moreover, we introduce E-PRCC and Occluded-VCCR, the first datasets dedicated to OCCRe-ID, providing new benchmarks for future research in this domain. In Video-based Occluded Re-ID (VOCCRe-ID), we first propose an occlusion-aware cross-attention fusion strategy, normalizing shape sequences using 3D SMPL models and applying spatio-temporal cross-modality learning to ensure consistency across frames. Finally, we introduce a bidirectional feature enhancement approach, where occlusions are synthesized progressively within sequences while keyframes remain unobstructed, enabling bidirectional feature propagation and correction. This method enhances robustness against both occlusions and clothing changes, ensuring temporal consistency in long-term person tracking. Furthermore, we construct Occluded-VCCR, the first dataset for VOCCRe-ID, aiming to facilitate further research on this practical task. Overall, this dissertation presents ten novel methods that significantly improve person Re-ID performance under real-world challenges. Through a combination of multi-modal learning, generative augmentation, cross-attention mechanisms, and occlusion modeling, this research contributes to the development of more reliable and scalable Re-ID systems for practical deployment
Factor Structure and Validation of a Complex PTSD Measurement for College Women
No full textRecently, researchers have identified a response to trauma, labeled complex posttraumatic stress disorder (CPTSD) which includes symptoms of posttraumatic stress disorder (PTSD) and disturbances in self organization (DSO; affect dysregulation, interpersonal difficulties, and negative self-concept). Purpose: The purpose of this study was to examine the construct validity of the Complex Trauma Inventory (CTI), an instrument to assess the core symptoms of PTSD and DSO that was developed with a college student sample in the United States. This study examined to what extent the factor structure of the CTI is replicated, and examined the convergent and predictive validity of the CTI subscale scores in relation to scores in other measures of PTSD, DSO, and mental health symptoms. Methods: Participants were asked to complete self-report measures describing their traumatic experiences and symptoms, functional impairment, and endorsement of depression and anxiety symptoms. Confirmatory factor analyses were used to examine two second-order factors (PTSD and DSO; Model 1) and six first-order factors (Model 2) models described by Litvin et al. (2017). Pearson’s correlational analyses were conducted to determine whether the CTI scores demonstrate convergent validity with PTSD and DSO symptoms assessed with the International Trauma Questionnaire (ITQ) and discriminant validity in relation to measures of anxiety and depression symptoms. Results: Results indicated that, as expected, the factor structure of the CTI was replicated. Confirmatory factor analyses provided support for the model suggested by Litvin et al. (2017), composed of six first order factors and two second-order factors (PTSD and DSO). These findings provide support to the theoretical understanding of DSO and PTSD as related, but distinct, symptom clusters. Convergent and predictive validity of the CTI subscale scores was also supported, with depression symptoms strongly correlating with DSO and both depression and anxiety symptoms with PTSD. Conclusion: The findings support previous structures and correlates of the CTI and offer a preliminary investigation into a measure that can continue to be developed with a college population in the United States
Understanding the Role of Adipokines in Cardiometabolic Dysfunction: A Review of Current Knowledge
No full textCardiometabolic risk and associated dysfunctions contribute largely to the recent rise in mortality globally. Advancements in multi-omics in recent years promise a better understanding of potential biomarkers that enable an early diagnosis of cardiometabolic dysfunction. However, the molecular mechanisms driving the onset and progression of cardiometabolic disorders remain poorly understood. Adipokines are adipocyte-specific cytokines that are central to deleterious cardiometabolic alterations. They exhibit both pro-inflammatory and anti-inflammatory effects, complicating their association with cardiometabolic disturbances. Thus, understanding the cardiometabolic association of adipokines from a molecular and signaling perspective assumes great importance. This review presents a comprehensive outline of the most prominent adipokines exhibiting pro-inflammatory and/or anti-inflammatory functions in cardiometabolic dysfunction. The review also presents an insight into the pathophysiological implications of such adipokines in different cardiometabolic dysfunction conditions, the status of adipokine druggability, and future studies that can be undertaken to address the existing scientific gap. A clear understanding of the functional and mechanistic role of adipokines can potentially improve our understanding of cardiovascular disease pathophysiology and enhance our current therapeutic regimen in the years to come
Nonlinear Optical Response of Organizational Chiral Surface and a Self-Assembled Monolayer on Silver Films as a Function of Film Thickness by Sum Frequency Generation Spectroscopy
No full textSurfaces and interfaces act as the boundary between a material and its surrounding environment, affecting interactions with that environment. To fundamentally understand the underlying processes, knowing the chemical and physical properties of these surfaces or interfaces is crucial. The Vibrational Sum Frequency Generation Spectroscopy (SFG) is a unique nonlinear optical process that provides vibrational spectra of molecules directly from the interface. The use of a metallic layer as a substrate is very usual in the SFG measurement which is employed to establish a non-vibrationally resonant reference against which the phase of resonant response can be compared. As the phase of the metal is significantly influenced by variables affecting the electronic structure of the interface, such as chemical bonding, surface coverage, metal film thickness, and the wavelength of visible light, it is advantageous to analyze the phase response from metal since the spectral line shapes are greatly affected by the nonresonant term. Here it is shown that the conventional line shapes for alkanethiol in the methyl stretching region changed significantly, with resonance appearing as dips being transformed into peaks with various silver film thicknesses. Moreover, SFG is an exceptional tool to explore the optical response from an organized chiral surface where molecules are organized in a chiral fashion on a substrate surface. This experimental work presents a phase shift and a direct assessment of the relative phase and amplitude alteration of a commonly utilized hydrophobic self-assembled monolayer (SAM) of 1-Octadecanethiol (ODT) adsorbed to various thicknesses of silver films as well as the SFG response from an organized chiral surface under different polarization of lights
The Impact of Daily Partner Gratitude on Intimate Partner Violence and Negative Mood during the COVID-19 Pandemic
No full textRates of intimate partner violence (IPV) have increased globally during the COVID-19 pandemic and virtual interventions may be useful for couples endorsing relational aggression. This project examines the relation between the impact of the COVID-19 pandemic, internalizing symptoms and IPV. Further it tests the efficacy of a daily gratitude text-based intervention for couples endorsing relational aggression. In a sample of 85 heterosexual couples, Actor Partner Interdependence Modeling (APIM) revealed that the negative impact of COVID-19 reported by men was significantly linked to their own physical IPV perpetration (actor effect), and women’s reported IPV perpetration (partner effect). Negative COVID-19 Impact reported by women was associated with their partner’s perpetration of psychological IPV (partner effect). However, women’s reported negative COVID-19 impact was not associated with their own physical IPV perpetration or their own psychological IPV perpetration. Although it was initially predicted that depression and anxiety would mediate the relation between COVID-19 stress and IPV, neither APIM was mediated by internalizing symptoms. Examining the daily diary data, one’s negative feelings on a given day were significantly associated with increased rates of relational aggression on the following day. Examining the effect of the gratitude intervention, those randomly assigned to daily gratitude exercises did not differ from those who were not in terms of daily negative mood or in partner aggression over the course of 30 days, as compared to the comparison condition. This study provides useful clinical information on both the mechanisms of how COVID-19 stress affects IPV and the efficacy of a cost-effective, accessible relationship intervention
Exploration of the Workings of Neural Networks
No full textWe study simple neural network architectures to gain a better understanding of their internal mechanisms. Firstly, we are developing a basic gradient-descent algorithm to train a single-layer neural network. We are deriving the first- and second-order derivatives and implementing a Gauss-Newton method for numerical optimization. We evaluate the performance of our approach using the MNIST dataset and compare it with pre-existing implementations of more intricate neural networks. Additionally, we are analyzing the performance of pre-existing dense neural networks in approximating simple functions. We report results for the discrepancy between the network's prediction and the true function based on the network's parameters and the function's complexity.Mathematics, Department ofHonors Colleg
Stroke Risk Stratification and Prevention Using Machine Learning and Explainable AI
No full textIn 2021, stroke was the fifth leading cause of death in the United States. Yet, research suggests that up to 80% of strokes could have been prevented. To better predict and prevent strokes, we are creating an XAI model capable of predicting stroke and coming up with alternative situations where stroke risk is lowered. These alternative scenarios serve to provide users with preventative measures they can take, as well as insight into the decision-making process of machine learning models. Specifically, we hope that the increased transparency from XAI will allow healthcare professionals greater confidence when using our model. To create our best-performing model, we pooled data from the 2001-2021 BRFSS dataset to artificially inflate stroke incidents in our dataset. Then, we selected 18 features that we found to be most important when classifying stroke cases, whilst excluding the features that may be affected by time, like income. Then, trained an AdaBoost model with a hyperparameter-tuned random forest as the base estimator. After training and testing, we found that our model achieves an accuracy of 71%, precision of 76%, recall of 71%, F1 of 71%, and ROC_AUC of 83%. Combining this model with Diverse Counterfactual Explanations (DiCE), we then generate alternative scenarios, local feature importance, and global feature importance. By locking unchangeable features, we can generate scenarios where users can greatly reduce their risk of stroke. This combination approach allows for a model that can help predict and prevent stroke, while also being more transparent.Biomedical Engineering, Department ofHonors Colleg
Does Religion Help with Nature Exposure? Faith as a Moderator During Summer Camp
No full textBackground: The positive youth development (PYD) movement utilizes a strengths-based approach in which specific conditions can be facilitated to promote social, emotional, cognitive, behavioral, and moral/spiritual growth that contributes to further successful lifelong development amongst adolescents (Park & Peterson, 2006). A prominent PYD model proposed by Lerner and colleagues (2005) is the Five Cs model of youth development – Competence, Confidence, Connection, Character, and Caring. According to this model targeting the 5 Cs promotes increases in development assets (DA) resulting in enhanced subjective well-being (SWB) and eudemonic well-being (EWB) as well as reduced mental health symptomology (Vitterso, 2016; Gillham et al., 2011). This suggests a dynamic and reciprocal link between DA, SWB/EWB, and mental health-promoting behaviors, which has been noted for camp experiences (Garst et al., 2016). In summer camps, PYD may be enhanced by outdoor experiences and exposure to nature, both of which are linked with cognitive, behavioral, and spiritual changes including recovery from fatigue, facilitating social cohesion, opportunities for physical activity, stress reduction, improved cognition, lower risk of depression/anxiety and enhanced spiritual experiences (Clearly et al., 2017; Chang et al., 2020). Such findings suggest that individuals may derive greater wellbeing from experiences in nature due to mechanisms associated with spiritual fulfillment (Trigwell et al., 2014). Daily R/S experiences are strong predictors of both life satisfaction, positive affect, and meaning in life or EWB (Yoon et al., 2014). Purpose: This may indicate that a plausible mechanism of action by which youth may differentially benefit from camp experiences is through heightened religiosity or spirituality (R/S) including greater religious attendance or self-reported religiosity which may lead to greater prosocial outcomes including character development. However, research on the moderating effect of R/S on outdoor adventure experiences for high school aged youth is limited. Methods: The current study was conducted on approximately 120 students who underwent a 2-week camping expedition with The Woods Project (TWP), most of whom were moderately religious (68.9%). R/S measured as a trait variable based on extrinsic and intrinsic religiosity were used as moderators to account for changes in pre- to post-trip outcome variables. Results: significant pre to post changes on 11 of the 12 outcome variables were found signifying that most students benefited from TWP. Regression analysis with age, ethnicity, and gender as covariates and extrinsic/intrinsic religiosity (separated) found mostly non-significant findings, such that religious/spiritual students did not differ significantly on the amount of change in outcome variables. Findings include one significant trend toward less growth and three non-significant trends toward greater growth associated with R/S. Importantly, the observed effects all moderators assessed, including R/S and race/ethnicity, were small and the study was underpowered to find small effects. Conclusion: Taken together, the study found minimal moderation of the robust pre-to postchanges observed. Further research on moderators of PYD program outcomes will need very large sample sizes to have sufficient power. Also, future studies should look at complex interactions around acculturation, identity formation, operationalization of religiosity/spirituality, and the role of secular and religious camping expeditions on positive youth development