11351 research outputs found
Sort by
Computational Learning Across Biological And Industrial Systems: Bayesian And Segmentation Models For Epigenomic And Manufacturing Data
No full textIn this thesis, we made a complete dual-domain investigation using a machine learning approach into two different scientific areas which were epigenetic analysis of seal species and pore identification in the work of additive manufacturing. The study showcased the flexibility and strength of modern computational tools in addressing complicated issues in various biological and industrial systems. We further investigated the epigenetics by using the Bayesian Neural Network and other machine learning methods to conduct a study on the DNA methylation pattern within three pinniped species (the northern elephant seal, Hawaiian monk seal, and Weddell seal) with perfect precision on species type identification and tissue origin differences. In the manufacturing field, we proposed U-SAMNet, an uncertainty-aware self-attention multi-task network, for the pore detection in Additive Manufacturing, conveying 99.83% accuracy and 91.11% F1-score in an efficient manner. The cross-domain comparison showed several shared challenges, such as the data imbalance, uncertainty quantification, and the requirement to design a robust pre-processing pipeline. In addition, this work introduced new methods to two different fields and it showed the potential translation of machine learning to scientific practice.
Index Terms: Additive manufacturing, Bayesian neural networks, DNA methylation, epigenetics, machine learning, multi-task learning, uncertainty quantification
(SI15-121) Analyzing SEITR Tuberculosis Transmission Model using Caputo–Fabrizio Fractional Derivative with Diverse Contact Rates
No full textIn the modern age, tuberculosis remains a pressing global health concern. Our study introduces and evaluates the SEITR pandemic TB transmission model, dividing the population into five compartments to explore distinct characteristics relevant to our investigation. Additionally, we delve into the application of fractional calculus. Through the Laplace transform method, we derive series solutions for all compartments, ensuring their existence and uniqueness. We also investigate the reproduction number of the tuberculosis epidemic model, examining how varying contact rates impact disease spread. We apply the predictor-corrector method for the Caputo-Fabrizio fractional derivative to verify the accuracy of our approach. This accurately predicts how the disease will change over time and makes our model more accurate. Our findings provide insight into the mechanisms underlying tuberculosis disease and show how the Caputo-Fabrizio fractional derivative can improve the accuracy of disease modeling
Emotional Pathways To Trauma: Racialized Policing And Psychological Distress In Black Adolescent Males
No full textIntroduction: Black American males, including adolescents, are disproportionately subjected to aggressive policing, often without criminal wrongdoing (Geller et al., 2014; Sewell et al., 2016). These encounters are frequently described as traumatic, marked by fear, loss, and emotional disruption (Smith Lee & Robinson, 2019). Whether direct or vicarious, such experiences can produce lasting psychological effects, particularly among Black boys who face compounded vulnerability from racialized surveillance and community-level trauma (DeVylder et al., 2018; Smith & Patton, 2016). Proactive policing tactics, such as investigatory stops and verbal aggression, have been linked to symptoms consistent with post-traumatic stress disorder (PTSD), including hypervigilance, avoidance, and emotional dysregulation (Bor et al., 2018; Bryant-Davis et al., 2017). Despite increased attention to police violence, emotional mechanisms underlying non-lethal encounters remain underexplored. Method: This study integrated Critical Race Theory and Vile Vigilance into a mediation model assessing emotional responses to policing. Using secondary data from the Fragile Families and Child Wellbeing Study (FFCWS), the sample included 365 Black males ages 15–18 (M = 15.74, SD = .746). Measures included self-reported police contact, emotional responses such as anger, fear, perceived safety, and trauma symptoms. Results: Investigatory stops and verbal aggression were significantly associated with increased trauma symptoms. Fear and anger mediated the effects of investigatory actions, while verbal aggression maintained a direct impact on trauma beyond emotional mediation. Use of force did not significantly predict trauma when emotional distress was accounted for, suggesting psychological impact may hinge more on perception than physicality. Conclusion: Findings reinforce prior research linking law enforcement actions to psychological distress and offer new insight into how emotional responses shape trauma outcomes among Black boys (McLeod et al., 2022). This study underscores the psychological toll of racialized policing and highlights the need for trauma-informed practices, culturally competent mental health care, and community-based strategies to mitigate harm and foster healing.
Keywords: African American youth, African American males, mental health, law enforcement contact, anxiety, depression, post-traumatic stress, law-enforcement-community relation
Implementing A Nurse Navigator Role In An Outpatient Procedural Setting To Improve Patient Satisfaction Scores: A Quality Improvement Project
No full textMuch focus has been placed on patient experience since the Affordable Care Act was signed into law in 2010 due to its regulatory, financial, and business growth implications. Patient satisfaction is crucial to maintaining and growing a customer base and fulfilling the target growth established in institutional strategic plans. This project aimed to determine whether implementing a nurse navigator role to perform rounds in the preoperative and waiting areas for two months would increase patient-reported satisfaction scores regarding being informed about delays and the likelihood of recommending the facility. Due to its subjective nature, it is challenging to consistently maintain high levels of patient satisfaction. Current approaches to improving patient satisfaction, including intentional hourly rounding by staff, are well-documented in inpatient settings; however, the effectiveness of these interventions in ambulatory, procedural, or surgical settings remains unclear. The PICO(T) question this project addressed was: Does a nurse navigator conducting rounds in the waiting area and preoperative rooms increase Press Ganey satisfaction scores regarding (a) being informed about delays during the visit and (b) the likelihood of recommending the facility compared with pre-intervention survey results over two months? Third-party vendor Press Ganey sent out 653 surveys, and 136 were completed and returned, resulting in a 20.8% return rate. The respondents were comprised of 62.5% males and 37.5% females, and 79.9% identified as White, while 16.5% identified as Black/African American. The Chi-square test of independence was used to compare pre- and post-implementation data. The nurse navigator intervention significantly improved patient scores regarding being informed about delays (p = .019). The change in the patients’ likelihood to recommend the facility was not significant (p = .887), indicating that the nurse navigator did not have a significant effect on this outcome. As patient satisfaction is multifaceted, a nurse navigator intervention should not be a facility’s sole strategy to increase patient satisfaction scores.
Keywords: nurse navigator, patient satisfaction, outpatient satisfaction, same-day discharg
Vegetation Health Prediction Using RGB Images From Uav With Deep Learning
No full textVegetation health prediction is one of the most important research questions in precision agriculture. The Normalized Difference Vegetation Index (NDVI) is the most widely adopted indicator of vegetation health, such as crop vigor and stress. Traditional NDVI estimation relies on multispectral or hyperspectral imaging, which is expensive and inaccessible for small-scale farming. Recent advances in deep learning enable the prediction of NDVI from low-cost RGB imagery; however, the limited availability of paired RGB-NDVI datasets and domain discrepancies between existing sources hinder robust generalization. In this thesis, we constructed UAV-based RGB-NDVI paired datasets and investigated multiple deep learning architectures, including a Fully Connected Neural Network (FCNN), an RGB-Split Convolutional Neural Network (RSCNN), and an Autoencoder with Dual Loss (ADL). Our experimental results showed that the RSCNN consistently delivered the strongest regression performance across datasets, achieving R2 values of 82.43% and 89.53%, and the lowest MAE of 1.87% and 1.57% on Dataset 1 and Dataset 2, respectively. Furthermore, we studied transfer learning and proposed a self-learning approach that relied on pseudo-labeled target samples. The results demonstrated improved performance that varied with the chosen confidence threshold and the transfer direction.
Index Terms: ADL, deep learning, FCNN, hyperspectral imagery, multispectral imagery, NDVI, precision agriculture, RSCNN, self-learning, transfer learning
(R2135) System Dynamical Analysis for ANN-Based Numerical Solutions of a Compartmental Model: A Bio-Mathematical Model of Drug Diffusion through the Compartments of Blood and Tissue
No full textThis paper provides a considerably efficient numerical approach to acquire the solutions of a biomathematical model administrating oral and intravenous distribution of pharmaceuticals in the human body. The proposed numerical approach based on an artificial neural network is employed to extract numerical solutions for a detailed set of ordinary differential equations and analyze the change in concentration of drug diffusion via the compartments of blood and tissue medium. We primarily focus on analyzing three different models established on the diffusion process, exercising laws of mass action and Fick’s principle. In this work, the existing model is reformulated as an optimization problem by investigating the drug distribution impacted by multiple factors related to the human body. Based on the drug efficacy, the rate constants (governing the law of mass action) are applied at different interfaces. The posed optimization problem is then solved by minimizing the concerned loss function. Also, all the attached numerical parameters have been considered while computing the drug concentration within distinct compartments. In addition, with the aid of Python programming, the presented plots show how the medication concentration changes over time. The obtained graphical results signify that the rate of change in the concentration of drugs rises gradually in other compartments while decreasing in the first. Compared to the traditional methods, the experimental results demonstrate the accuracy and efficacy of the proposed methodology evidently
Comparative Evaluation Of Power And Timing Simulation-Based Side Channels For Hardware Trojan Detection
No full textHardware Trojans are covert modifications to integrated circuits that alter function or leak information while avoiding traditional verification. This thesis presented a simulation-based side-channel study of the AES-T1800 benchmark, utilizing power and timing analyses. The methodology aligned value change dump signals with waveform windows to correlate internal switching with external power signatures and compared timing slack distributions between a Trojan Free build and a Trojan Intruded build after implementation in the design tool.
Power analysis revealed a clear and repeatable fingerprint during Trojan activation. In filtered traces, the Trojan Free run rose from about 0.8×10⁻⁷ W to a steady level near 4.0×10⁻⁷ W with minimal short-term variability, while the Trojan Intruded run exhibited transient peaks and oscillations before settling with peaks near 5.0×10⁻⁷ W, which was roughly 20% to 25% above the steady baseline. These peaks aligned with windows where Trojan control and payload registers toggled in the simulation, providing a direct causal link between added internal switching and elevated instantaneous power and spectral energy. The uploaded graphs reflected these differences and supported automated metrics such as delta energy, transition count difference, and spectral contrast. Timing analysis quantified the effect on path delays but with lower practical sensitivity. For the same measured path set, the Trojan Free build showed a mean slack of ≈+1.020 ns with 0 failing paths and a minimum slack of ≈+0.024 ns. The Trojan Intruded build showed a mean slack of ≈−2.973 ns, worst slack of ≈−6.142 ns, and 20 failing paths. These results indicate a substantial regression in setup timing while hold timing remained safe, consistent with added combinational delay and longer nets.
The comparative conclusion from these timing and power studies is that both modalities indicated the presence of AES-T1800, but, power provided the stronger and more reliable detection signal under realistic variation and noise. The power method yielded a higher signal-to-noise ratio through localized, short-duration spikes and elevated energy, exactly coincident with Trojan activity, whereas timing shifts were smaller and more easily masked without strict environmental control. In this study, power was the superior primary method, and timing serves as a complementary check.
Index Terms—Advanced Encryption Standard (AES), simulation side channel, Trojan detection
(SI14-03) Time Proportional Non-Instantaneous Deterioration Decisions for Vendor Managed Inventory System
Full text linkThis article formulates a model for optimizing inventory replenishment decisions for the products that experience non-instantaneous deterioration by allowing partial backlogging. Two-level supply chain (SC) is considered in vendor-managed inventory (VMI) model which contains ‘single-retailer and single-supplier’ for studying the performance. The model considers two constant demand rates which differ from others when deterioration in the product commences. In addition, the general function of time is considered for deterioration rate. Theoretical results are derived to compute the replenishment cycle length. Lastly, the benefit of VMI policy is illustrated through numerical examples. Further, explanation of some features of the VMI model is made using sensitivity analysis for presenting the liability of this model in the form of managerial results
(SI14-11) Analysing Co-current Imbibition Phenomenon in Heterogeneous Reservoir using Multistep Hybrid Differential Transform Finite Difference Method
Full text linkThe primary focus of this study is to analyse the co-current imbibition phenomenon in an inclined heterogeneous reservoir. This phenomenon occurs during the secondary oil recovery process. Capillary force is responsible for the displacement of a non-wetting phase by a wetting phase, and this phenomenon is called spontaneous imbibition. Imbibition is of two types and can be differentiated based on the direction in which the wetting phase (water) and non-wetting phase (oil) move. If the two phases flow in the same direction, it is called co-current imbibition, and if they flow in the opposite direction, it is called counter-current imbibition. In the oil recovery process, oil and water form the two phases that are immiscible, with water being the wetting phase. Here, water enters from one end while oil leaves from the other end. The partial differential equation that arises from the mathematical formulation of this phenomenon is non-linear, and obtaining an exact solution is tedious or sometimes difficult. The Multistep Hybrid Differential Transform Finite Difference Method combines the multistep differential transform and finite difference methods to obtain the solution. This is a semi-analytic numerical approach that gives the solution in the form of an infinite series. The solution determines the saturation of injected water at various distances and time levels. Additionally, how the angle of inclination impacts the saturation of injected water is examined. MATLAB is used to obtain the numerical solution and generate graphical representations